Iterative Imputation of Incomplete Wireless Network Traffic using Adversarial Learning and a Two-Stage Approach

In modern industries, wireless communication technologies are the driver of flexible information transfer. To meet certain communication requirements like reliability, low latency and a specific data rate, constant monitoring and analysis of the captured data is necessary. Passive monitoring is a technique for tracking the communication between devices on link-level. Under certain conditions, the monitoring device loses information (data packets) from time to time. To avoid erroneous data analysis results, incomplete traces have to be reconstructed by imputation. In the present work, we propose a deep learning based imputation approach, using the concept of Generative Adversarial Networks (GANs). For our experiments we created a simulated data set using ns-3. The proposed model could outperform the selected baseline methods. We further improved the result by implementing a second stage of imputation based on the Expectation Maximization (EM) algorithm.